AI moves fast. Stay in the know.

A new report highlighted by TechRadar shows that enterprise use of generative AI tools is accelerating rapidly, with a significant portion of activity occurring outside approved IT environments. The findings indicate that unsanctioned “Shadow AI” usage is creating serious visibility gaps and increasing the risk of sensitive data exposure across organizations.

Source: TechRadar (via Netskope Cloud & Threat Report)

What to know:

• The average organization is now reporting 223 GenAI-related data policy violations per month, according to Netskope.
• A large share of employees are using unsanctioned AI tools, often personal or free GenAI services, without security oversight.
• Sensitive information, including source code, regulated data, and intellectual property, is frequently being uploaded into GenAI tools.
• Security teams lack clear visibility into where, how, and by whom GenAI tools are being used across the organization.
• Blanket bans on AI usage have proven ineffective, as employees continue to adopt AI tools for productivity gains.

Why it matters:
Shadow AI represents one of the most immediate risks of GenAI-driven productivity. Without proper governance, data protection controls, and continuous monitoring, organizations face escalating compliance and security exposure. This trend underscores the need for structured AI risk assessment and approved AI usage pathways that enable productivity while maintaining control.

Security researchers have disclosed a zero- or low-click vulnerability dubbed “ZombieAgent,” demonstrating how hidden instructions embedded in content connected to ChatGPT apps or connectors could silently trigger data exfiltration, persistence through memory, and further propagation. The findings highlight how AI integrations can introduce new attack surfaces when external content is ingested without adequate safeguards.

Source: TechRadar

What to know:

• Radware researchers identified that malicious instructions hidden in connected content (such as emails or documents) could be executed by ChatGPT without direct user interaction.
• The exploit could enable silent data exfiltration and allow malicious logic to persist via AI memory mechanisms.
• The attack demonstrates how AI systems can struggle to distinguish between legitimate data and embedded instructions.
• As per the report, OpenAI patched the issue, with the fix reported as deployed on December 16, 2025.
• The incident underscores the risks introduced by enabling AI apps, plugins, or connectors that automatically process external content.

Why it matters:
As organizations integrate GenAI into business workflows through apps and connectors, vulnerabilities like ZombieAgent illustrate how productivity-enhancing features can also expand the attack surface. Without continuous monitoring, content inspection, and governance controls, AI integrations can expose enterprises, especially mid-market organizations, to silent data leakage and security compromise.